Can not understand the results by using ghost atom

[zhangguangping]

Dear ATK developer and ATK user,

I use ATK to investigate the distance between graphene and the Cu(111) surface. In the calculation, there are five layers of Cu(111) with the leftmost three layers fixed and the two surface layers optimzed.  During the optimization, the graphene is also relaxed. I used GGA functional and a SZP basis set for Cu atoms/DZP for C atoms. The result shows that the distance between graphene and the surface is only 2.4 Angstrom, which is much smaller to that of 3.5 Angstrom reported by other studies. So, I used a layer of ghost atom to do the optimization again, where the ghost atom layer is also optimized. The optimized positions of ghost atoms are very close to the surface layer. In this way, I would expect a very similar distance between graphene and the surface. However, the optimized distance is 3.38 Angstrom, which is close to the ones in literatures. How to understand it?

With best regards,

Guang-Ping Zhang

[Jess Wellendorff]

A ghost atom is nothing but a local basis set positioned some place in space. The DZP/SZP basis set you use is relatively short-ranged (which is sometimes very good), but may not be accurate enough for the study you do. Adding the ghost atom appears to "enlarge" the effective basis set range, which changes your result. I suggest you try to use the default SG15 pseudopotentials and basis sets, which should be both more accurate and longer ranged.

[zhangguangping]

A ghost atom is nothing but a local basis set positioned some place in space. The DZP/SZP basis set you use is relatively short-ranged (which is sometimes very good), but may not be accurate enough for the study you do. Adding the ghost atom appears to "enlarge" the effective basis set range, which changes your result. I suggest you try to use the default SG15 pseudopotentials and basis sets, which should be both more accurate and longer ranged.

Dear Jess,

My question is that since the ghost atoms are very close to the surface atoms the basis set should not be essentially enlarged and a very similar result should be expected to  that without ghost atoms. But the results are very different. You mean both my results are unreliable?

With best regards,

Guang-Ping Zhang

[Daniele Stradi]

Dear Guang-Ping Zhang

in graphene-M(111), with M being a noble metal, the interaction between the graphene and the metal is mediated mainly by charge transfer to/from the metal surface state.
It is very likely that, by including the ghost atom, you (i) improve the description of the Cu(111) surface state, and (ii) reduce somehow the effect of the basis set superimposition error at typical binding distances.

For relevant literature, see:
Phys. Rev. B 79, 075441
Phys. Rev. B 94, 155431   

Regards,
Daniele.

[zhangguangping]

Dear Guang-Ping Zhang

in graphene-M(111), with M being a noble metal, the interaction between the graphene and the metal is mediated mainly by charge transfer to/from the metal surface state.
It is very likely that, by including the ghost atom, you (i) improve the description of the Cu(111) surface state, and (ii) reduce somehow the effect of the basis set superimposition error at typical binding distances.

For relevant literature, see:
Phys. Rev. B 79, 075441
Phys. Rev. B 94, 155431   

Regards,
Daniele.

Dear Daniele,

Thanks a lot for your reply.

With best regards,

/Guang-Ping Zhang